NVIDIA’s New Moat Building Strategy
These days, it seems that if NVIDIA needs something to be secured, it writes a US$2 billion cheque. Synopsys in December. CoreWeave in January. Lumentum and Coherent on the same March Monday. Nebius the following week. Marvell days ago. The pattern is so metronomic it resembles less a portfolio strategy than an automated irrigation system: a US$2 billion sprinkler rotates slowly across the AI supply chain, depositing a standard round wherever the ground looks dry.
The jokes write themselves. But the mechanism underneath is consequential when one focuses on the aggregate pattern. What NVIDIA is running is no longer a corporate venture program. It is not behaving like a venture capitalist with a GPU. It resembles a procurement department with a stockbroker. The deals used to be secure revenues with the help of compute credits before; now they are supply reservations that happen to trade on Nasdaq.
The distinction matters because it reframes everything: the economics of the instrument, the competitive logic, the accounting treatment, and above all, the question of who else will be forced to follow. NVIDIA's public equity portfolio has grown from roughly US$230 million two years ago to over US$13 billion in disclosed holdings by end-2025, with the March 2026 wave adding another US$8-10 billion. These investments could become a regular affair, like other forms of Capex, not just for the GPU giant but even for others.
First, the Most Obvious: Supply Wars
The supply chain logic is straightforward, even if bizarre, as historically the struggle has always been for market share and locking the biggest clients. Component providers and other types of vendors have historically attracted little attention. For most of the last decade, cloud credits, subsidized compute bills, and customer lock-in were the instruments designed for a world where filling capacity was the hard problem. The vendors were always kept nervous, with the threat of moving the business to the next-door neighbor as the primary partnership tool.
By now, even the most resistant previous-era tech observers agree that the binding constraint is everything upstream: coherent laser components, silicon photonics, advanced packaging, EDA software, custom ASICs. We are not even mentioning semiconductor manufacturing, which is perhaps the mother of all, except that those players are unlikely to go for strategic investments (with one exception in our eyes that we discussed here).
And once you understand the logic that way, the question is not why NVIDIA is deploying US$2 billion per deal into listed suppliers. The question is what else is likely in the quarters ahead.
Many commentators have already reached for the Japanese keiretsu analogy: a network of cross-shareholding industrial firms coordinated by a main bank and a trading house. There are surface similarities. But the more important framing is that NVIDIA is building a new type of moat: not a product moat, not a software moat, but a supply-chain moat that competitors cannot easily replicate without years of relationship-building, capital deployment, and technical integration unless they enter the fray fast. ASML's 2012 experience, when Intel, TSMC, and Samsung co-invested to fund EUV development, is the closer precedent. That has been the best-known example of customers using equity to de-risk a shared, irreplaceable supplier. NVIDIA is now doing it solo and systematically across at least seven categories simultaneously.
As we wrote in our note on the real chip wars, the ASIC competitors to NVIDIA are not small. They are also entities with enormous cash piles and active corporate strategy departments. One wonders how long before we see announcements from others.
Queue Stratification, Not Capacity Reservation
The conventional framing of these deals is supply-chain security. That framing is correct, but it undersells the instrument. There are multiple approaches here: the two most discussed in the semiconductor space are long-term agreements (being used by memory-makers) and precommitments, often with prepayment (most popular with TSMC). A long-term supply agreement is tremendously valuable as it provides supply-chain security, even with price risks. A prepayment is even stronger. The reason equity outperforms both comes down to geometry: it creates a shared destiny rather than a contractual obligation.
A supply agreement alone is only as strong as the supplier's incentive to honor it. Without capital alignment, a supplier can always find a higher-margin customer willing to pay more in a tight market. A prepayment is better, but it sits as a receivable on the investor's balance sheet, provides no upside participation if the supplier executes well, and offers no inherent reason for the supplier to deepen the relationship beyond the payment term. Both are unilateral commitments from the buyer.
Equity changes the geometry entirely. If the joint technology development succeeds and the supplier's stock rises, the investor participates. If the relationship sours, the equity position is liquid (at least in theory, and subject to a critical caveat we will return to). The equity is also a commitment device that makes the accompanying purchase commitment credible in both directions. The supplier is incentivized to prioritize NVIDIA's needs, knowing it has a long-term capital partner. NVIDIA is locked into a reliable source of supply, insulated from competitors' poaching capacity. The combination of equity stake plus purchase commitment creates something neither instrument can achieve alone.
In the old economy, price determined priority. In this one, balance sheet intimacy may. Equity contribution is effectively queue stratification. When Coherent or Lumentum has to choose which customer's order to accelerate in a constrained month, the customer who owns 4-5% of your equity and has committed to multi-billion dollar forward purchases is not in the same queue as everyone else. No contract may say that explicitly because none needs to.
From the buyer, NVIDIA’s viewpoint, any buybacks with excess cash would retire some stock. These stakes retire uncertaint. In supply constrained world, the most accretive use of cash may not be buying your own equity at 28 times earnings, but buying your place in someone else's production schedule.
Not Bothersome to the Usual Hawks
Accounting purists may be among the most satisfied observers of this turn of events. Under US GAAP ASC 321, minority stakes in publicly listed companies are marked to fair value through the income statement each quarter, with no election or smoothing available. A US$2 billion position moving 30% generates US$600 million of quarterly noise for a company running quarterly operating income in the tens of billions, proportionately modest, and above all, transparent. Unlike an impairment on a private position that can be disguised behind "measurement alternative" accounting for years, a public equity mark-to-market signals clearly when a relationship is deteriorating. Plus, the investments are not made to generate revenues that weren’t; now they are to ensure that genuine revenue business is completed.
Antitrust regulators present a more nuanced picture. The individual deals are defensible: minority stakes below 5%, no board seats, no governance rights, no formal exclusivity in the supply agreements. But, in the US, the 2023 Merger Guidelines introduced a serial acquisition doctrine under Guideline 8 that is genuinely novel: a firm engaging in a pattern of small acquisitions across related supply chains can face scrutiny even when no single deal is problematic. NVIDIA has now taken minority stakes in seven listed companies, all within the same AI infrastructure stack. No enforcement action has targeted this pattern. But the statutory framework exists, and a future administration or a single determined commissioner could start targeting these deals if they become more commonplace or lead to anti-monopoly type complaints from its competitors.
Industrial policy advocates and national security observers will, if anything, view this trend with enthusiasm bordering on relief. The investments in Lumentum and Coherent, like Intel before, each explicitly fund new US manufacturing capacity. The capital is functioning as a private industrial policy, accelerating onshoring without requiring a government grant or a subsidy that invites political controversy. The CHIPS Act can fund fabs; only a committed strategic customer-shareholder can guarantee the demand that makes those fabs economically viable. NVIDIA is simultaneously de-risking the supply chain for its own products and reducing US dependence on offshore optical component manufacturing. The Nokia investment fits the same frame: a domestic AI-RAN ecosystem built around a trusted Western vendor.
For researchers and economists studying inter-firm integration, this wave is arguably the most interesting natural experiment in vertical coordination since the Japanese keiretsu studies of the 1980s, albeit with a crucial methodological advantage. Because these are publicly listed companies with real-time price discovery, mark-to-market accounting, and mandatory SEC disclosure, the depth and health of the relationships can be observed continuously rather than reconstructed from surveys or internal documents. The one caveat that accounting and legal scholars will flag is the MNPI problem: For now, if a purchaser is buying a stake in a business with the knowledge of its own demand, it rightly does not generate attention. The reverse will not be easy. In times of severe downturn, the stakeholders of the acquiring companies, like NVIDIA, may want derisking of the portfolio or more cash, but the sale of assets may not be easy for the company, knowing more than other investors in the market.
Frankly, there will be numerous MNPI versus fiduciary duty issues likely in the coming years. Just imagine the dilemma if NVIDIA’s internal team develops a method that destroys the business of one of its investments. It could be sued by shareholders for not working in the best interest, if it does not sell, and surely, it could be sued by everyone else, including the regulators, if it does.
Not All Suppliers Take Water
The model works elegantly when the supplier genuinely needs the cheque. Lumentum was funding a new US fabrication facility. Coherent needed balance sheet to expand its domestic manufacturing footprint. Nebius needed credibility as much as capital. Nokia needed repositioning more than it needed cash. In each case, NVIDIA's equity arrived at a genuine moment of capital need, which is precisely why the alignment is real and the commitment credible in both directions. The sprinkler irrigates the plants that are thirsty. It does nothing useful for the ones that are not.
Which brings us to TSMC, Samsung Memory, and SK Hynix. Each generates cash at a rate that dwarfs any US$2 billion minority cheque NVIDIA could write. None has a capital need that a strategic investor could fill. Each would gain less from the relationship signal than it would lose in governance complexity and customer-perception problems. When the Trump administration explored requiring TSMC to cede equity in exchange for CHIPS Act subsidies, TSMC held preliminary discussions about returning the subsidies. NVIDIA's purchase commitments with TSMC are among the largest in the foundry's history. The relationship does not need equity to be binding.
AMD's mirror-image approach introduces a different set of tensions. By issuing warrants to OpenAI and Meta at a nominal US$0.01 exercise price, each representing up to 10% of AMD's equity, vesting against gigawatt-scale deployment milestones, AMD is using its own stock as a loyalty instrument rather than deploying cash. The commercial logic is elegant. The governance implications for AMD's existing shareholders are less so. If both warrant packages vest fully, dilution approaches 20% with no cash consideration flowing to AMD beyond the purchase commitments. Shareholders who held AMD before these deals were struck are effectively underwriting the economics of customer relationships they did not vote on and cannot easily exit. The AMD board may have made a rational business decision. Whether it made a fair one for pre-existing holders is a question that will attract more attention if the warrants move toward vesting thresholds.
The Arrival of the Corporate Fund Managers
NVIDIA is the most visible practitioner, but the underlying logic is spreading. The two catalysts are related: the deteriorating economics of private equity exits, and the extreme concentration of cash in a small number of listed corporates.
The most enthusiastic capital deployers in late-stage venture rounds by 2021 were not independent funds but corporate treasuries. The pattern has been so strong that it even began attracting complaints from dedicated investors who found themselves crowded out by strategics without return discipline. That CVC wave has not reversed, but it has slowed and in several sectors, particularly application-layer software, the moats proved shallower than the capital committed. Private exit channels remain constrained. The IPO window has reopened unevenly. And the most strategically important companies in AI infrastructure are already public.
Meanwhile, the corporate world has become structurally more unequal in cash terms. A handful of companies, and not just in the US, generate cash at a rate that no historical precedent quite captures, while facing antitrust constraints that effectively cap their ability to make large acquisitions. That combination of enormous cash generation, limited acquisition runway, acute supply constraints, and rising competition for the same bottleneck resources is precisely the environment in which public equity becomes the obvious instrument.
The pattern of adoption is already wider than the NVIDIA headlines suggest. In March 2025, Amazon issued warrant arrangements with both Fabrinet and Applied Optoelectronics tied to purchase commitments. Fabrinet, whose revenue is approximately 35% driven by NVIDIA, received warrants giving Amazon the right to purchase 381,922 shares at US$208.49. Applied Optoelectronics issued warrants to Amazon covering up to 7.95 million shares, vesting against US$4 billion of discretionary purchases. Neither arrangement generated anything close to the attention of NVIDIA's March 2026 deals, which is precisely the point: the instrument was already being adopted quietly, in a form that requires no large upfront cash commitment and creates supply alignment without balance sheet exposure. In January 2026, Eli Lilly anchored Aktis Oncology's IPO with US$100 million. Lilly is not seeking a financial return on US$100 million of Aktis stock. It is ensuring that a platform technology central to its oncology pipeline has the capital to reach clinical proof of concept, and signaling to potential acquirers that Aktis comes with a strategic attachment. The form differs from NVIDIA's model. The function is nearly identical.
Notwithstanding the Eli Lilly deal above, the pharmaceutical sector is moving in this direction through non-equity mechanisms. Cross-border manufacturing deals, milestone-linked clinical collaborations, and capacity reservation agreements tied to clinical success are proliferating rapidly in recent months. These are not equity stakes, but they reflect the same recognition: that supply of a critical input is worth paying to secure before the constraint becomes acute. The non-equity version of the model may actually be more durable in this sector than pure equity because of the likely regulatory sensitivity if the purchased stake is high and the ineffectiveness if the purchased stake is trivial.
The Boardrooms Are Already Burning
The story does not end with what NVIDIA has done. It begins there. What happens next is already in motion in ways that do not yet appear in press releases.
Start with the fear that is not being stated publicly. When Lumentum and Coherent each received a US$2 billion cheque, the most consequential conversations likely took place in the corporate strategy offices of Google, Amazon, Microsoft, and a few others. Each of those companies runs operations that directly or indirectly depend on the same optical components NVIDIA just acquired preferential access to. Of course, none said anything publicly. All of them must be asking internally what it would take to establish a comparable arrangement with the suppliers they have not yet approached.
The CUDA analogy must be circulating. The supply chain moat NVIDIA is now building is the physical-world equivalent: not a software dependency but a capital dependency, a commitment dependency, a queue dependency.
Analysts covering the optical networking space have documented that hyperscaler procurement teams are struggling to secure transceiver and laser capacity on timelines that were previously routine. Until now, Jensen Huang has been building relationships that others envied with frequent Asia tours and fried chicken dinners. Now, the game has changed far more. If one were learning from Apple’s premium customer label loss at TSMC, the deals are a more solid signal of what may happen ahead if they do not act.
So far, Amazon appears to be the only one aware of the power of these deals based on the transactions we mentioned above.
The geographic extension is also coming, and the first movers in non-US markets will have an advantage that is not yet being priced. Japan's photonics and advanced materials companies, Korea's packaging specialists, and European precision optics names: none of these have yet seen a systematic approach from a strategic investor, but we will not be surprised if the conversations have started. The regulatory overlay is real — Japan's FEFTA, Korea's FDI screening, CFIUS for inbound US investment — but manageable at sub-5% stakes held as genuinely passive positions. The first large US corporate to take a meaningful minority stake in a listed Japanese advanced packaging company or a listed Korean substrate manufacturer will generate more than the announcement-day premium for the securities involved.
In all likelihood, what NVIDIA has started this March is not a coincidence, a fad, or even a trend. This is a structural realignment of how large companies may manage the relationship between their balance sheets and their supply chains, driven by forces — corporate cash concentration, antitrust limits on acquisitions, physical supply constraints, and intensifying competition for the same bottleneck resources — that are not going away. In a supply-constrained world, that distinction is not at the margin. It is the entire game.
